| """
|
| JaneGPT v2 Model Architecture
|
|
|
| A lightweight decoder-only transformer with classification head
|
| for intent classification. Features modern architecture components:
|
| - Rotary Position Embeddings (RoPE)
|
| - Grouped Query Attention (GQA)
|
| - SwiGLU feed-forward networks
|
| - RMSNorm
|
|
|
| Created by Ravindu Senanayake
|
| """
|
|
|
| import math
|
| import torch
|
| import torch.nn as nn
|
| import torch.nn.functional as F
|
| from typing import Optional, Tuple
|
|
|
|
|
|
|
| INTENT_LABELS = [
|
| "volume_up", "volume_down", "volume_set", "volume_mute",
|
| "brightness_up", "brightness_down", "brightness_set",
|
| "media_play", "media_pause", "media_next", "media_previous",
|
| "browser_search", "app_launch", "app_close", "app_switch",
|
| "set_reminder", "screenshot", "read_screen", "explain_screen",
|
| "undo", "chat", "quit_jane",
|
| ]
|
| INTENT_TO_ID = {label: i for i, label in enumerate(INTENT_LABELS)}
|
| ID_TO_INTENT = {i: label for i, label in enumerate(INTENT_LABELS)}
|
| NUM_INTENTS = len(INTENT_LABELS)
|
|
|
|
|
| class RMSNorm(nn.Module):
|
| """Root Mean Square Layer Normalization."""
|
| def __init__(self, dim, eps=1e-6):
|
| super().__init__()
|
| self.weight = nn.Parameter(torch.ones(dim))
|
| self.eps = eps
|
|
|
| def forward(self, x):
|
| rms = torch.sqrt(torch.mean(x ** 2, dim=-1, keepdim=True) + self.eps)
|
| return (x / rms) * self.weight
|
|
|
|
|
| class RotaryEmbedding(nn.Module):
|
| """Rotary Position Embeddings (RoPE)."""
|
| def __init__(self, head_dim, max_seq_len=512, theta=10000.0):
|
| super().__init__()
|
| inv_freq = 1.0 / (theta ** (torch.arange(0, head_dim, 2).float() / head_dim))
|
| self.register_buffer('inv_freq', inv_freq)
|
| t = torch.arange(max_seq_len).float()
|
| freqs = torch.outer(t, inv_freq)
|
| self.register_buffer('cos_cached', torch.cos(freqs))
|
| self.register_buffer('sin_cached', torch.sin(freqs))
|
|
|
| def forward(self, seq_len):
|
| return self.cos_cached[:seq_len], self.sin_cached[:seq_len]
|
|
|
|
|
| def apply_rope(x, cos, sin):
|
| """Apply rotary position embeddings to input tensor."""
|
| head_dim = x.shape[-1]
|
| x1 = x[..., :head_dim // 2]
|
| x2 = x[..., head_dim // 2:]
|
| rotated = torch.cat([-x2, x1], dim=-1)
|
| cos = cos.unsqueeze(0).unsqueeze(0).repeat(1, 1, 1, 2)
|
| sin = sin.unsqueeze(0).unsqueeze(0).repeat(1, 1, 1, 2)
|
| return x * cos + rotated * sin
|
|
|
|
|
| class GroupedQueryAttention(nn.Module):
|
| """
|
| Grouped Query Attention (GQA).
|
|
|
| Uses fewer KV heads than query heads for memory efficiency
|
| while maintaining attention quality.
|
| """
|
| def __init__(self, embed_dim, num_heads, num_kv_heads, head_dim,
|
| max_seq_len, dropout, rope_theta):
|
| super().__init__()
|
| self.num_heads = num_heads
|
| self.num_kv_heads = num_kv_heads
|
| self.head_dim = head_dim
|
| self.num_groups = num_heads // num_kv_heads
|
|
|
| self.q_proj = nn.Linear(embed_dim, num_heads * head_dim, bias=False)
|
| self.k_proj = nn.Linear(embed_dim, num_kv_heads * head_dim, bias=False)
|
| self.v_proj = nn.Linear(embed_dim, num_kv_heads * head_dim, bias=False)
|
| self.out_proj = nn.Linear(embed_dim, embed_dim, bias=False)
|
|
|
| self.dropout = nn.Dropout(dropout)
|
| self.rope = RotaryEmbedding(head_dim, max_seq_len, rope_theta)
|
|
|
| def forward(self, x):
|
| batch_size, seq_len, _ = x.shape
|
|
|
| q = self.q_proj(x).view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
|
| k = self.k_proj(x).view(batch_size, seq_len, self.num_kv_heads, self.head_dim).transpose(1, 2)
|
| v = self.v_proj(x).view(batch_size, seq_len, self.num_kv_heads, self.head_dim).transpose(1, 2)
|
|
|
| cos, sin = self.rope(seq_len)
|
| q = apply_rope(q, cos, sin)
|
| k = apply_rope(k, cos, sin)
|
|
|
| if self.num_groups > 1:
|
| k = k.repeat_interleave(self.num_groups, dim=1)
|
| v = v.repeat_interleave(self.num_groups, dim=1)
|
|
|
| scores = (q @ k.transpose(-2, -1)) / math.sqrt(self.head_dim)
|
| mask = torch.triu(torch.ones(seq_len, seq_len, device=x.device), diagonal=1).bool()
|
| scores = scores.masked_fill(mask, float('-inf'))
|
|
|
| attn_weights = torch.softmax(scores, dim=-1)
|
| attn_weights = self.dropout(attn_weights)
|
|
|
| out = attn_weights @ v
|
| out = out.transpose(1, 2).contiguous().view(batch_size, seq_len, -1)
|
| return self.out_proj(out)
|
|
|
|
|
| class SwiGLUFeedForward(nn.Module):
|
| """SwiGLU Feed-Forward Network."""
|
| def __init__(self, embed_dim, ff_hidden, dropout=0.1):
|
| super().__init__()
|
| self.w1 = nn.Linear(embed_dim, ff_hidden, bias=False)
|
| self.w2 = nn.Linear(ff_hidden, embed_dim, bias=False)
|
| self.w3 = nn.Linear(embed_dim, ff_hidden, bias=False)
|
| self.dropout = nn.Dropout(dropout)
|
|
|
| def forward(self, x):
|
| return self.dropout(self.w2(F.silu(self.w1(x)) * self.w3(x)))
|
|
|
|
|
| class TransformerBlock(nn.Module):
|
| """Single transformer block with GQA and SwiGLU."""
|
| def __init__(self, embed_dim, num_heads, num_kv_heads, head_dim,
|
| ff_hidden, max_seq_len, dropout, rope_theta):
|
| super().__init__()
|
| self.norm1 = RMSNorm(embed_dim)
|
| self.norm2 = RMSNorm(embed_dim)
|
| self.attn = GroupedQueryAttention(
|
| embed_dim, num_heads, num_kv_heads, head_dim,
|
| max_seq_len, dropout, rope_theta
|
| )
|
| self.ff = SwiGLUFeedForward(embed_dim, ff_hidden, dropout)
|
|
|
| def forward(self, x):
|
| x = x + self.attn(self.norm1(x))
|
| x = x + self.ff(self.norm2(x))
|
| return x
|
|
|
|
|
| class JaneGPTv2Classifier(nn.Module):
|
| """
|
| JaneGPT v2 Intent Classifier.
|
|
|
| A decoder-only transformer with a classification head
|
| for 22-class intent classification.
|
|
|
| Args:
|
| vocab_size: Vocabulary size (default: 8192)
|
| embed_dim: Embedding dimension (default: 256)
|
| num_heads: Number of attention heads (default: 8)
|
| num_kv_heads: Number of KV heads for GQA (default: 4)
|
| num_layers: Number of transformer layers (default: 8)
|
| ff_hidden: Feed-forward hidden dimension (default: 672)
|
| max_seq_len: Maximum sequence length (default: 256)
|
| dropout: Dropout rate (default: 0.1)
|
| rope_theta: RoPE theta parameter (default: 10000.0)
|
| """
|
| def __init__(self, vocab_size=8192, embed_dim=256, num_heads=8,
|
| num_kv_heads=4, num_layers=8, ff_hidden=672,
|
| max_seq_len=256, dropout=0.1, rope_theta=10000.0):
|
| super().__init__()
|
|
|
| self.embed_dim = embed_dim
|
| self.max_seq_len = max_seq_len
|
|
|
| self.token_embedding = nn.Embedding(vocab_size, embed_dim)
|
| head_dim = embed_dim // num_heads
|
|
|
| self.layers = nn.ModuleList([
|
| TransformerBlock(
|
| embed_dim, num_heads, num_kv_heads, head_dim,
|
| ff_hidden, max_seq_len, dropout, rope_theta
|
| )
|
| for _ in range(num_layers)
|
| ])
|
| self.norm = RMSNorm(embed_dim)
|
| self.dropout = nn.Dropout(dropout)
|
|
|
| self.intent_head = nn.Sequential(
|
| nn.Linear(embed_dim, embed_dim),
|
| nn.GELU(),
|
| nn.Dropout(dropout),
|
| nn.Linear(embed_dim, NUM_INTENTS),
|
| )
|
|
|
| def forward(self, x, labels=None):
|
| x = self.dropout(self.token_embedding(x))
|
| for layer in self.layers:
|
| x = layer(x)
|
| x = self.norm(x)
|
|
|
| pooled = x[:, -1, :]
|
| logits = self.intent_head(pooled)
|
|
|
| loss = None
|
| if labels is not None:
|
| loss = F.cross_entropy(logits, labels)
|
|
|
| return logits, loss
|
|
|
| @torch.no_grad()
|
| def predict(self, x):
|
| logits, _ = self.forward(x)
|
| probs = F.softmax(logits, dim=-1)
|
| confidence, predicted = torch.max(probs, dim=-1)
|
| return predicted.item(), confidence.item() |
